ABEILLE: a novel method for ABerrant Expression Identification empLoying machine LEarning from RNA-sequencing data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Omics technologies advance rare disease diagnosis by identifying causative genes.
• Transcriptomic data analysis for aberrant gene expression (AGE) can reveal pathogenic events.
• Current AGE identification methods rely on arbitrary statistical cut-offs and require multiple replicates, which are often unavailable in clinical settings.

Purpose of the Study:

• To develop a novel machine learning-based method for identifying aberrant gene expression (AGE) from RNA-seq data.
• To overcome limitations of existing AGE identification approaches, specifically the need for replicates and control groups.
• To provide a flexible and robust tool for identifying potential pathogenic genes in rare disease research.

Main Methods:

• Developed ABerrant Expression Identification empLoying machine LEarning from sequencing data (ABEILLE), a variational autoencoder (VAE)-based method.
• Utilized a VAE to model RNA-seq data without distributional assumptions, combined with a decision tree for gene classification.
• Incorporated an anomaly score to stratify identified aberrant genes by severity.

Main Results:

• ABEILLE successfully identified aberrant gene expression (AGE) from RNA-seq data without requiring replicates or a control group.
• The method demonstrated flexibility in VAE configuration for identifying potential pathogenic candidates.
• Performance was validated on both semi-synthetic and experimental datasets.

Conclusions:

• ABEILLE offers a powerful, flexible, and data-efficient approach for identifying aberrant gene expression in rare disease contexts.
• The VAE-based methodology overcomes key limitations of traditional statistical methods for AGE detection.
• ABEILLE facilitates the discovery of novel genetic factors in rare diseases through advanced transcriptomic analysis.